Panel 4

Producing Knowledge with/about Machines (in english)

Chair: Paula Helm (Tübingen)

Recently in educational theory and psychology, a focus has been placed on Augmented Reality (AR), as with it comes the “promise of creating direct, automatic, and actionable links between the physical world and electronic information” (Schmalstieg and Hollerer 2016), thereby increasing overall learning performance. In this talk, however, I am primarily interested whether and how employing AR might shift the kind of knowledge that is acquired and what possible repercussions this may have. 

Drawing on the longstanding epistemological debate about the differences between Knowledge-How (KH) and Knowledge-That (KT), the talk explores different ways in which AR environments change understanding through visualizations, trial-and-error learning, and gamification elements. As a result, using AR in education may contribute to a shift in knowledge acquisition from a focus on factual knowledge to a focus on competencies, and thereby facilitate the ongoing paradigm shift in educational theory from encyclopedic factual knowledge to teaching competencies and soft skills.

In various academic disciplines, work is increasingly being done with new data sources, such as social media content or data traces generated by the use of Internet-of-Things technologies. The new research methods of data science and machine learning are used in the analysis of this data, and are also often applied on social media platforms to analyze user behavior, reflect trends or even predict them. New types of knowledge production emerge from complex collaborations between human and non-human actors. The research meant here is embedded in a network of connections between the variable affordances of (commercial), more-or-less automated and algorithmically curated internet platforms; the third-party tools that support data collection, processing, and analysis; different data formats; research institutions and infrastructures; and the tools and platforms provided by research communities themselves. Who acts as originator of certain data or content is no longer necessarily apparent or traceable. A traditional source criticism, for example, would quickly reach its limits. 

The contribution will examine the new collaborations between human and non-human actors in the generation of knowledge. The focus will be on how these collaborations are both globally networked and locally anchored.

In recent years, the field of data-driven decision support has developed rapidly due to significant advances in the area of machine learning. This development has opened up new opportunities in a variety of social, scientific, and technological fields. However, it is becoming increasingly clear from experience, particularly in the social sphere, that focusing on purely statistical and numerical aspects in data analysis fails to capture social nuances or take ethical criteria into account. A widespread assumption is that data-based software systems can replace people in their decision-making processes. Often, the introduction of software is seen as a “substitution problem.” In a fixed human workflow, specific tasks are replaced by an algorithm, which results in, among other things, less work, fewer errors, and higher accuracy. However, two challenges need to be considered: Technologies are not value-neutral. Delegating formerly manual tasks to software (or vice versa) leads to significant shifts in social practices and responsibilities. 

Therefore, in this talk, I advocate an alternative approach when designing software systems. By using participative methods, existing interdependencies of collaborative human-machine activities should be disclosed. The insights gained from these methods can complement quantitative machine learning approaches with qualitative data. Furthermore, the software’s decision model should be transparently communicated and interactively explorable by humans.